Detailed table of the prediction results
Information about the match prediction model for UEFA 2007-2008
The amount of factors that influence the outcome of a football match is of course immense and collecting data on all of them would be very difficult. The processing of all this data in an intelligent way would also be near impossible. But by making use of easily accessible data, BayesIT has developed a knowledge model that predicts outcomes of the matches surprisingly well. It uses an unique technology, consisting of so called Bayesian Networks, structure that is a sort of probabilistic calculus. This predicts better than a lot of other technologies, especially in cases with sparse, or missing data.
We tried to find suitable latent variables using a technology demo during the World Cup 2006. The results were very encouraging. BayesIT offers now for football fans a joint prediction model for the UEFA CL and Cup. It is intended for the season 2007-8 but the knowledge model can be used for any prediction as long as the user recognizes the importance of the historical data that has been used to teach the model.
About the prediction model
Following factors have been taken into account in the knowledge model:
- Both UEFA Champions League and Cup match results are used from the past. Eliminations are considered partly.
- From season 2004-5 onwards all matches are used. For 2001-2 only final matches, for 2002-3 only final stage matches.
- All CL and Cup matches played during season 2007-8 to date have been considered.
- Intertoto matches are considered for seasons 2006 and 2007.
- The number of goals in each match.
- The number of seasons the team has participated in previous UEFA CL and Cups is partly considered.
- The distance in points to the previous years' champion team.
- The FIFA ranking of some participant teams are considered but only as indications.
- The FIFA ranking of participant teams‘ countries are considered but only as indications.
- The teams’ home stadium capacity.
The modelling algorithm also takes into account the difference between the prediction about the historical match and the actual result in the data. For the sake of this model, we assume that the performance of a team that does not conform to the prediction of the model is a consequence of temporary influences such as morale boosts from winning streaks, setbacks from injuries etc. All of which might make the team either over- or underperform.
Weighting of more recent behaviour of the teams is replaced by modelling
Differing from our previouse football prediction model, the gradually diminishing impact of the matches played in the distant past to the overall team performance is learned through the modelling. It's not a weighting, as the learning algorithm decides by itself, how much importance it gives to the various factors.
The development of the model continues. You can read more about the underlying bayesian method here.
The UEFA 2007-2008 prediction model was created using BayMiner
BayMiner is a browser-enabled tool for analysis of data in tabular format. It finds dependencies in complex multi-dimensional situations and visualizes their co-occurrences in a format easily understandable by humans. The data does not need to be numeric and the source table does not need to be complete. Read more on frequently asked questions on BayMiner.
Using BayMiner for business
Understanding complex situations and timely reaction to changing situations is a common theme in business. Instead of classical statistics producing multitude of obtuse curves and pie-charts about a particular set of data, BayMiner uses machine-learning to produce an easy to use knowledge model. BayMiner makes the structure of even very complex data-sets easily accessible for the user. Unlike conventional data mining tools, BayMiner is very easy to use and does not require knowledge about statistics or the technology used.
Benefits of Bayesian Networks
Bayesian Networks are the best technology for mastering uncertainty. They make possible to grasp complex cause-dependencies in multi-dimensional environments such as quality disturbances, or project management. One of the reasons for using Bayesian Networks in this football application, as well as in many business applications, is that a knowledge model using Bayesian Networks can be successfully used for predictions when a time series is not long enough to use classical statistical (frequentistic) methods.
Working in a modelled environment gives you the opportunity to make trustworthy decisions quickly, in response to various needs as they arise. The fast implementation of knowledge models is a major factor behind the success of BayMiner in many applications.
BayMiner allows better analysis of latent variables
Latent variables are variables that are not directly observable but may be inferred from, usually several unknown other variables that are observed and measured. It is not always possible to identify and name a latent variable, as it sometimes represents a very complex phenomenon. Latent variables are also called hidden variables. Examples in football context include “fighting spirit” and in business context would include such aspects as “confidence” or “early adobtability”.
Examples of successful business applications
The BayMiner solution is widely acknowledged as the most flexible solution available for business analysis today. With its probabilistic modelling suite BayMiner PRO, the dedicated applications EWS (Early Warning Signals) and QVM (Quality Variance Management), BayesIT has the technology and services that fit literally any need of a business organization.
EWS
Knowledge models based on data about operations in the past and developed in BayMiner can predict future more accurately than most other technologies. For the purpose of simplicity of use, BayesIT developed the BayMiner EWS (Early Warning System) product. EWS applications are usable with rudimentary skills with using computers. This spearhead application steers the user using traffic lights away from risky tenders in an early stage of project or contract service business. For more information, See our pages on EWS.
QVM
Probabilistic Modelling solutions coupled with 3D visualization help users identify the connection between e.g. quality and process parameters. For this purpose BayesIT developed the QVM concept. It is especially useful in detecting root causes to claims and punctuality variations (delays in production). For more information, See our pages on QVM.
Other applications
Significant benefits in various business applications can be achieved, including:
- Reduced costs through identification of root causes to quality problems.
- Reduced customer dissatisfaction through identification of root causes to field problems.
- Definition of right customer service levels using multi-dimensional segmentation.
- Improved profitability through identification of non-profitable feature and market combinations.
- Improved delivery performance through identification of excessive order information changes.
- Fraud detection.
Go to the predictor
